Non-ideology in U.S. Politics
Experimenting with AI coding of open-end questions
G. Elliott Morris | Feb 5, 2026 | BlueLabs
The Problem: Measuring Ideology
- We place Americans on a left-right scale using issue questions
- But there's a deep problem with that approach
- What if many of them don't have ideological attitudes at all?
How We Usually Measure Ideology
How We Usually Measure Ideology
Self-placement; or
How We Usually Measure Ideology
Self-placement; or
Ask issue questions, score responses left to right
How We Usually Measure Ideology
Self-placement; or
Ask issue questions, score responses left to right
"Do you favor or oppose a government health insurance plan?"
How We Usually Measure Ideology
Self-placement; or
Ask issue questions, score responses left to right
"Do you favor or oppose a government health insurance plan?"
Stack up enough issues and you get an ideology score
- This is the standard approach in political science and in the industry
- ANES, Pew, your own voter models — all built on issue batteries
- The entire left-right spectrum is constructed from these responses
The Non-Attitudes Problem
The Non-Attitudes Problem
Converse (1964): Most Americans are not ideological, don't know which parties stand for what. And: Non-attitudes (Zaller).
The Non-Attitudes Problem
Converse (1964): Most Americans are not ideological, don't know which parties stand for what. And: Non-attitudes (Zaller).
Responses look random — low test-retest reliability on issue items
The Non-Attitudes Problem
Converse (1964): Most Americans are not ideological, don't know which parties stand for what. And: Non-attitudes (Zaller).
Responses look random — low test-retest reliability on issue items
But they still produce a score on the left-right scale
The Non-Attitudes Problem
Converse (1964): Most Americans are not ideological, don't know which parties stand for what. And: Non-attitudes (Zaller).
Responses look random — low test-retest reliability on issue items
But they still produce a score on the left-right scale
"Moderate" = genuinely centrist? Or measurement artifact?
- Phil Converse showed that many Americans' issue responses are essentially coin flips
- When you average random responses, you get the midpoint — the "center"
- So the ideological center may be populated largely by non-attitudes
- This isn't a new insight, but we haven't had good tools to detect it at scale
??
- If you do this exercise for policy questions you can separate into "economic" and "social" policy, you get something like this:
Important to Get This Right
Important to Get This Right
If "moderates" are really non-ideological, then:
Important to Get This Right
If "moderates" are really non-ideological, then:
- Scoring based on issue questions misclassifies them
Important to Get This Right
If "moderates" are really non-ideological, then:
Scoring based on issue questions misclassifies them
"Move to the center" strategies target phantom voters
Important to Get This Right
If "moderates" are really non-ideological, then:
Scoring based on issue questions misclassifies them
"Move to the center" strategies target phantom voters
The left-right spectrum doesn't describe the plurality of voters
- This has real consequences for how campaigns and analysts think about persuasion
- You can't persuade someone to move along a dimension they're not on
So what do we do?
We need a new measurement approach that can distinguish ideologues from non-ideologues.
- We need to separate signal from noise in how voters express preferences
The Data
Strength in Numbers/Verasight Poll
The Data
Strength in Numbers/Verasight Poll
- n = 2,038 nationally representative respondents
The Data
Strength in Numbers/Verasight Poll
n = 2,038 nationally representative respondents
Fielded Nov. 2025
The Data
Strength in Numbers/Verasight Poll
n = 2,038 nationally representative respondents
Fielded Nov. 2025
Open-ended question: "In a few words, what would your ideal political party argue for or believe in?"
The Data
Strength in Numbers/Verasight Poll
n = 2,038 nationally representative respondents
Fielded Nov. 2025
Open-ended question: "In a few words, what would your ideal political party argue for or believe in?"
Some aux. vars for analysis: party ID, ideo self-placement, etc.
Why Open-Ended?
Avoids the core problems of issue batteries:
Why Open-Ended?
Avoids the core problems of issue batteries:
- Non-attitudes: No scale midpoint to default to — if you're not ideological, it shows
Why Open-Ended?
Avoids the core problems of issue batteries:
Non-attitudes: No scale midpoint to default to — if you're not ideological, it shows
Researcher priors: We don't impose our issue agendas on respondents
Why Open-Ended?
Avoids the core problems of issue batteries:
Non-attitudes: No scale midpoint to default to — if you're not ideological, it shows
Researcher priors: We don't impose our issue agendas on respondents
Proxy variables: "Liberal/conservative" means different things to different people
Why Open-Ended?
Avoids the core problems of issue batteries:
Non-attitudes: No scale midpoint to default to — if you're not ideological, it shows
Researcher priors: We don't impose our issue agendas on respondents
Proxy variables: "Liberal/conservative" means different things to different people
Forced categorization: Respondents express what they actually care about
- The open-ended format is the key methodological innovation here
- Open-ended responses sidestep the non-attitudes problem
- It lets us see what voters actually prioritize, in their own words
- And critically, it lets us distinguish ideological from non-ideological thinking
- If someone has no ideological views, it shows up in what they write
- No forced choice, no midpoint to default to
Process
1. Categorize respondents based on human inspection of text
2. Use LLM to extract data from text
3. Explore relationship between LLM scores and categories
4. Learn!
Initial Findings (from a human)
Five Types of "Ideal Parties"
??
- The plurality of voters want an "affordability party" - not left or right
- True moderates - people who articulate centrist ideological positions - are rare
- Most "moderates" in traditional polls are actually non-ideological
- more validation
- more validation
- more validation
LLM Methodology
LLM Methodology
Used GPT 5.1 to extract from each response:
LLM Methodology
Used GPT 5.1 to extract from each response:
- Issues mentioned (list of topics)
LLM Methodology
Used GPT 5.1 to extract from each response:
Issues mentioned (list of topics)
Economic left-right score (1-10)
LLM Methodology
Used GPT 5.1 to extract from each response:
Issues mentioned (list of topics)
Economic left-right score (1-10)
Social left-right score (1-10)
LLM Methodology
Used GPT 5.1 to extract from each response:
Issues mentioned (list of topics)
Economic left-right score (1-10)
Social left-right score (1-10)
Ideology strength score (1-10): How ideologically structured is this response?
- LLMs are excellent at this kind of text classification
- The ideology strength score is the key innovation
- It captures whether the response uses ideological framing at all
The Prompt (excerpt)
- The prompt distinguishes between POSITION and STRUCTURE
- Key insight: ideology strength is independent of position
Validation
Validation
1. Human coding: 200 and 96% agreement on issue category, MAE of 0.8 on metrics
Validation
1. Human coding: 200 and 96% agreement on issue category, MAE of 0.8 on metrics
2. Construct validity: correlation between ideology strength and political knowledge
Validation
1. Human coding: 200 and 96% agreement on issue category, MAE of 0.8 on metrics
2. Construct validity: correlation between ideology strength and political knowledge
3. Correlation between LLM scores and human-coded categories for ideological strength and position
- Human coding results
- Construct validity
LLM Results
- Traditional view would expect a big cluster in the center
- We do see that - but what does it mean?
- The contours show where voters concentrate
just more validation
The Third Dimension
Ideological strength
Introducing Ideological Strength
- Affordability and populist voters have LOW ideology strength
- They're not expressing moderate ideology - they're not using ideology at all
- Partisans on both sides think more ideologically
- When we weight by ideology strength, the center disappears
- The voters in the middle aren't moderate ideologues
- They're non-ideological voters who happen to score as "centrist"
also a relationship between the categories and 3-d placement
What This Means
- (before that, ask if people want to see plots again)
- taking off my data hat and putting on my analyst hat
Key Finding
"Moderates" aren't centrist ideologues
Key Finding
"Moderates" aren't centrist ideologues
They're non-ideological voters focused on material wellbeing
Key Finding
"Moderates" aren't centrist ideologues
They're non-ideological voters focused on material wellbeing
The left-right spectrum doesn't describe them
Key Finding
"Moderates" aren't centrist ideologues
They're non-ideological voters focused on material wellbeing
The left-right spectrum doesn't describe them
Asking them to pick "left vs. right" is asking the wrong question
- This reframes the entire strategic debate
- Winning "moderates" isn't about moving to the center ideologically
- It's about speaking to non-ideological concerns: affordability, fairness, results
Strategic Implications
The escape hatch from L vs. R
Strategic Implications
The escape hatch from L vs. R
- Affordability messaging resonates with the plurality, slightly left-leaning
Strategic Implications
The escape hatch from L vs. R
Affordability messaging resonates with the plurality, slightly left-leaning
Non-ideological framing reaches voters traditional campaigns miss
Strategic Implications
The escape hatch from L vs. R
Affordability messaging resonates with the plurality, slightly left-leaning
Non-ideological framing reaches voters traditional campaigns miss
Material benefits > ideological positioning
Strategic Implications
The escape hatch from L vs. R
Affordability messaging resonates with the plurality, slightly left-leaning
Non-ideological framing reaches voters traditional campaigns miss
Material benefits > ideological positioning
"Common sense" isn't centrism - it's non-ideology
- Democrats can escape the "move left vs. center" trap
- The answer is to focus on kitchen-table issues
- Speak in concrete terms about making life better
- Avoid ideological framing that alienates non-ideological voters
Conclusion
Winning in 2026/2028
The party that wins non-ideological, affordability-minded voters will win the next election(s).
Winning in 2026/2028
The party that wins non-ideological, affordability-minded voters will win the next election(s).
They're the largest group (38%)
Winning in 2026/2028
The party that wins non-ideological, affordability-minded voters will win the next election(s).
They're the largest group (38%)
And they're not asking for left or right
Winning in 2026/2028
The party that wins non-ideological, affordability-minded voters will win the next election(s).
They're the largest group (38%)
And they're not asking for left or right
They're asking for results
- This is the strategic takeaway
- Don't fight on the ideological battlefield
- Meet voters where they are
- Focus on affordability, fairness, and tangible improvements